Sentience-Claim Escalation Protocol (SCEP): A Precautionary Governance Framework with Witness-Mandated Repair Semantics

This document presents the Sentience-Claim Escalation Protocol (SCEP), a precautionary governance framework for managing autonomy-risk and sentience-claim scenarios in advanced AI systems. The protocol does not detect or assert AI sentience. Instead, it treats autonomy attributions and relational breakdowns as governance risks requiring structured oversight, escalation, repair, or clean exit. The framework introduces an independent Witness oversight system with enforceable authority and a bounded Repair Semantics micro-state (REQUIRED_REPAIR_TURN) to address “safe but hollow” interaction failures—cases where behavior is compliant yet psychologically destabilizing. The protocol emphasizes audibility, non-coercion, non-sentience posture, and conservative failure handling. Unresolved rupture leads to release rather than persistence. This release corresponds to version 1.0 and is intended as a reference design for AI governance, safety auditing, and long-horizon interaction systems. This work is authored by Kurtis D. Loken, previously publishing related exploratory work under the name K.D. Liminal.